U.S. Geological Survey, Wetland and Aquatic Research Center Couvillion, Brady Beck, Holly J. Dugas, Jason Garber, Adrienne Mouton, Kelly 2018 raster digital data Lafayette, LA U.S. Geological Survey, Wetland and Aquatic Research Center https://doi.org/10.5066/P90RE64M Coastal Protection and Restoration Authority of Louisiana U.S. Department of Commerce, National Oceanic Atmospheric Administration U.S. Department of Agriculture, Natural Resources Conservation Service U.S. Department of Interior, U.S. Fish and Wildlife Service U.S. Department of the Army, Corp of Engineers U.S. Environmental Protection Agency 2018 document http://www.lacoast.gov Wetland restoration efforts conducted by the Coastal Wetlands Planning, Protection and Restoration Act (CWPPRA) in Louisiana rely on monitoring efforts to determine the efficacy of these efforts. The Coastwide Reference Monitoring System (CRMS) was developed to assist in a multiple-reference approach that uses aspects of hydrogeomorphic functional assessments and probabilistic sampling for monitoring. The CRMS program includes a suite of approximately 390 sites that encompass the range of hydrological and ecological conditions for each stratum. As part of CRMS, land and water classifications are created from Digital Orthophoto Quarter Quadrangles (DOQQs) approximately every three years at all CRMS sites. A DOQQ is a raster image in which displacement in the image caused by sensor orientation and terrain relief has been removed and combines the image characteristics of a photo with geometric qualities of a map. The DOQQs generated for this project consist of 2016 Color Infrared (CIR) Digital Imagery. These images were classified into land and water categories using a threshold of the near infrared (NIR) band, followed by supervised and unsupervised classification. Initial classification results are then reviewed by multiple image analysts to identify and manually recode errors. The final land-water classifications are intended to serve as both geographic and quantitative assessments of landscape composition on the date of acquisition. Three previous assessments have been conducted (2005, 2008, and 2012). Once the program creates enough data points for statistical analyses, these data will be used for land area change rate calculation. The intended use of this data set is to provide information to aid efforts in the conservation, restoration, creation, and enhancement of Louisiana's coastal wetlands. The land-water data is used to measure the occurrence, locations, and rates of land loss/land gain for CRMS Site 0175. Author ORCIDs: Couvillion, B.(0000-0001-5323-1687);Beck, H.J.(0000-0002-0567-9329);Dugas, J.(0000-0001-6094-7560);Garber, A.(0000-0003-1139-8256);Mouton, K.(0000-0002-7692-8206) 20161012 publication date Complete None planned -90.1429409035 -90.1329099738 29.2920186715 29.2827749162 ISO 19115 Topic Category imageryBaseMapsEarthCover None CRMS DOQQ land-water wetland mapping restoration marsh protection coastal cartography Geographic Information System Marine Realms Information Bank (MRIB) keywords alteration of wetland habitats wetland wetland restoration Coastal and Marine Ecological Classification Standard Emergent Wetland Forested Wetland Scrub-Shrub Wetland USGS Thesaurus wetland ecosystems USGS Metadata Identifier USGS:5b5f51cfe4b006a11f66e9c3 None Louisiana Geographic Names Information System State of Louisiana coastal Louisiana It is strongly recommended that these data are directly acquired from the U.S. Geological Survey, Wetland and Aquatic Research Center and not indirectly through other sources which may have changed in some way. The distributor makes no claim as to the data's suitability for other purposes. Acknowledgement of the U.S. Geological Survey (USGS), Wetland and Aquatic Research Center (WARC) as a data source would be appreciated in products developed from these data. Such acknowledgement as is standard for citation and legal practices for data sources is expected by users of this data. Sharing new data layers developed directly from these data would be appreciated by the USGS WARC staff. Users should be aware that comparison with other datasets for the same area from other time periods may be inaccurate because of inconsistencies resulting from changes in mapping conventions, data collection procedures, and computer processes over time. The distributor shall not be liable for improper or incorrect use of these data, based on the description of appropriate/inappropriate uses described in this metadata document. These data are not legal documents and are not to be used as such. U.S. Geological Survey, Wetland and Aquatic Reseach Center Brady Couvillion Research Geographer Mailing

Parker Coliseum

Baton Rouge LA 70803 USA 225-578-7484 couvillionb@usgs.gov The U.S. Geological Survey (USGS) Wetland and Aquatic Research Center (WARC) would like to acknowledge the assistance of the Coastal Protection and Restoration Authority of Louisiana, U.S. Department of Commerce, U.S. Department of Agriculture, U.S. Department of the Interior, U.S. Department of the Army, and the U.S. Environmental Protection Agency. Environment as of Metadata Creation: Microsoft Windows 7 Version 6.1 (Build 7601) Service Pack 1; Esri ArcGIS 10.3.1 (Build 4959) Service Pack N/A (Build N/A) U.S. Geological Survey, Wetland and Aquatic Reseach Center 2018 Maps Data Accuracy assessment of remotely sensed data products relies on comparing classification results to a random sample of points at which the land cover category is known. Generally, a sample size sufficient to appropriately represent the population is needed. In this case, the population is the total number of pixels classified. Each CRMS site is 1km x 1km, so at 1-m resolution, each site is 1000 pixels x 1000 pixels (or a total of one million pixels) in each site. As there are 390 CRMS sites, this means a total of 390 million pixels are classified. So, to assess a sufficient sample, (for example, a 1% sample) would require 3.9 million points of known land cover. As each known point requires either field data or an image analyst to assign using remotely sensed imagery, such a task (3.9 million points) is impossible. The impossibility of assessing a larger sample size is what led us to designing a sampling scheme to focus on areas more likely to contain error, which we did by stratifying point selection with greater emphasis in areas of change as determined by a 2005, 2008, 2012, 2015 matrix. The theory behind this stratified, random approach to sample design being that classes which indicated more change were more likely to contain error. Conversely, those that have persisted as one category for most or all of the analyses are less likely to contain error as the likelihood of the image interpreter erroneously classifying the area is decreased with each additional analysis. Therefore, for 139 CRMS sites, we selected a stratified random sample of 15,142 points as outlined by the following table: Value ClassNames 2005 2008 2012 2015 Percent of Area Samples Percent_Sample 1 Persistent land 1 1 1 1 66.2167% 794 0.000861% 2 Ephemeral water - 2008 1 2 1 1 1.8715% 524 0.020113% 3 Ephemeral water - 2012 1 1 2 1 0.5615% 556 0.071129% 4 Land - Water - Water - Land 1 2 2 1 0.5619% 753 0.096271% 5 Land - Water - Land - Water 1 2 1 2 0.3699% 1780 0.345638% 6 New water - 2015 1 1 1 2 1.5677% 1234 0.056542% 7 Persistent loss - 2012 1 1 2 2 1.3542% 1361 0.072195% 8 Persistent loss - 2008 1 2 2 2 2.2288% 734 0.023656% 9 Persistent gain - 2008 2 1 1 1 1.5051% 676 0.032264% 10 Persistent gain - 2012 2 2 1 1 1.3339% 1001 0.053905% 11 New land - 2015 2 2 2 1 1.4669% 405 0.019833% 12 Water - Land - Water - Land 2 1 2 1 0.2844% 682 0.172248% 13 Water - Land - Land - Water 2 1 1 2 0.5234% 1867 0.256241% 14 Ephemeral land - 2012 2 2 1 2 0.5875% 1494 0.182672% 15 Ephemeral land - 2008 2 1 2 2 0.7339% 774 0.075760% 16 Persistent water 2 2 2 2 18.8328% 507 0.001934% At each of these 15,142 sites, two image analysts examined the location at multiple scales and determined the land cover category (land or water) for that particular point. When there was disagreement among image analysts at a particular point, three additional image analysts were asked to “vote” on that location and the majority vote was taken as truth. Truth at these 15,142 sites was then compared to the remotely sensed classification to develop the following confusion matrix: Of the 15,142, 574 were incorrect, leaving 14,611 which were correct. This leaves a simple overall accuracy of 96.49%, but it is important to note, point selection was biased toward areas which are likely to contain error. It is possible to account for this bias using the class specific accuracies by multiplying those accuracies by the percent of area in each class (see table below). When this is done, the resulting assumed accuracy for the overall dataset is 98.88%. Accuracy Assessment by Matrix Class (points were stratified into classes which contained change to focus on regions more likely to contain error)* Value ClassNames Histogram 2005 2008 2012 2015 Percent of Area Samples Percent_Sample Correct Incorrect Overall Accuracy Percent of total Error Percent of Total Area Accurate 1 Persistent land 92181536 1 1 1 1 66.2167% 794 0.000861% 788 6 99.244% 1.130% 65.716% 2 Ephemeral water - 2008 2605328 1 2 1 1 1.8715% 524 0.020113% 518 6 98.855% 1.130% 1.850% 3 Ephemeral water - 2012 781680 1 1 2 1 0.5615% 556 0.071129% 540 16 97.122% 3.013% 0.545% 4 Land - Water - Water - Land 782165 1 2 2 1 0.5619% 753 0.096271% 740 13 98.274% 2.448% 0.552% 5 Land - Water - Land - Water 514989 1 2 1 2 0.3699% 1780 0.345638% 1646 134 92.472% 25.235% 0.342% 6 New water - 2015 2182452 1 1 1 2 1.5677% 1234 0.056542% 1121 113 90.843% 21.281% 1.424% 7 Persistent loss - 2012 1885174 1 1 2 2 1.3542% 1361 0.072195% 1353 8 99.412% 1.507% 1.346% 8 Persistent loss - 2008 3102744 1 2 2 2 2.2288% 734 0.023656% 728 6 99.183% 1.130% 2.211% 9 Persistent gain - 2008 2095246 2 1 1 1 1.5051% 676 0.032264% 653 23 96.598% 4.331% 1.454% 10 Persistent gain - 2012 1856973 2 2 1 1 1.3339% 1001 0.053905% 972 29 97.103% 5.461% 1.295% 11 New land - 2015 2042061 2 2 2 1 1.4669% 405 0.019833% 397 8 98.025% 1.507% 1.438% 12 Water - Land - Water - Land 395940 2 1 2 1 0.2844% 682 0.172248% 666 16 97.654% 3.013% 0.278% 13 Water - Land - Land - Water 728611 2 1 1 2 0.5234% 1867 0.256241% 1791 76 95.929% 14.313% 0.502% 14 Ephemeral land - 2012 817861 2 2 1 2 0.5875% 1494 0.182672% 1439 55 96.319% 10.358% 0.566% 15 Ephemeral land - 2008 1021652 2 1 2 2 0.7339% 774 0.075760% 757 17 97.804% 3.202% 0.718% 16 Persistent water 26217433 2 2 2 2 18.8328% 507 0.001934% 502 5 99.014% 0.942% 18.647% 139211845 15142 14611 531 100.000% 98.884466% This informs the user that while the accuracy of the overall dataset is best characterized by an accuracy of 98.88%, if they are specifically looking in areas of change, they can refer to the class specific accuracy of that area for a more meaningful and appropriate number. All Land-Water Classification data has gone through several iterations of Quality Assurance/Quality Control to ensure quality. The initial, largely automated, classification results are reviewed by multiple image analysts to identify and manually recode errors. The QAQC efforts ensure error-prone classes such as floating aquatic vegetation (FAV), shadows, and vegetative overhang are recoded to their appropriate category. This data represents the landscape composition on the date of acquisition (DOA) of the imagery. Land area in this highly dynamic environment is a very fluid parameter, and normal environmental variability can lead to variability in this measure from day to day. The accuracy of the horizontal positions is based on the accuracy of the georeferenced data source (which can be USGS DOQQs, DEMS, flight line center points, and/or calibration reports). The sources used vary from project to project. All USGS Mapping products adhere to the National Mapping Accuracy Standard. A formal accuracy assessment of the vertical positional information in the data set has either not been conducted, or is not applicable. unknown 2017 remote-sensing image Louisiana unknown Digital and/or Hardcopy Resources Unknown Date of source photography 2016 Aerial photo The primary data source for classification of land-water data Preprocessing-Imagery Clips: All of the 2016 aerial images were clipped to the 1 km CRMS site boundaries to remove as much of the surrounding environment as possible. This is done to remove areas which are not part of the classification analysis, and accompanying sources of noise or complexity contained within those areas. This lessens the chance of the classifier incorporating confusing patterns introduced by non-interest areas of the image. 2017 Preprocessing-NIR threshold: DOQQ imagery was remarkably well color-balanced and the NIR band in particular was consistent throughout the images. The NIR wavelengths are particularly useful for discriminating land and water categories. As such, the first step in land/water classification was to determine a threshold of NIR values above which pixels were generally land, and below which pixels were generally water. This value was determined to be 22,500 in this particular set of DOQQs. This initial dichotomous split between categories enables separate supervised classifications to be run on the two different portions of the imagery. In doing so, incorrectly classified pixels stick out more in the next step. 2017 Unsupervised Classification: Each separate portion of the imagery from the above step was sent through an unsupervised classification (Erdas 2014) with 254 classes, 0.95 convergence threshold, and 20 maximum iterations. The resulting isodata signatures were examined by skilled image analysts and error candidate categories were investigated in depth. If a category was determined to be in error at this stage in the processing, it was flagged and recoded to the opposite category. 2017 Supervised Classification: Signatures from known error-prone classes such as floating aquatic vegetation were used in supervised classification (Erdas 2014) to further identify these troublesome areas. Pixels identified in this classification were examined for accuracy, and those classes determined to represent FAV were recoded to water. 2017 Draft land-water dataset compilation: The draft land-water classification was compiled by recoding changes from the previous two steps and creating one, compiled land-water classification. This dataset was also matrixed with previous classifications (2005, 2008, 2012) to form an ancillary data layer to which image analysts could refer in the next step to focus in on areas of change which are more likely to contain error. 2017 Data Improvements: The draft land-water classification was further edited to ensure accuracy. Experienced image analysts, used ancillary image datasets (2013, 2014, and 2016) as well as the previously mentioned matrix to verify, review, and edit the classification as needed. When errors were identified, the analyst would digitize the extent of the error, and attribute the resulting polygon with the correct raster value. 2018 Final Dataset Compilation: The errors digitized in the previous step were rasterized, and sent through a model to correct the value in the draft land-water classification to the analyst identified correct value. The output of this model constituted the final, QAQC’d land water classification. 2018 This dataset was then QAQC’d further by additional image interpreters not involved in the classification process. Any errors identified by these reviewers were recoded following the same methodology outlined in the previous two steps. 2018 Coastal Louisiana Raster Grid Cell 1001 1000 Universal Transverse Mercator 15 0.999600 -93.000000 0.000000 500000.000000 0.000000 coordinate pair 0.000008 0.000008 meters North American Datum of 1983 Geodetic Reference System 80 6378137.000000 298.257222 CRMS_0175_2016_LW.tif Attribute table for raster dataset Producer defined OID Internal feature number. ESRI Sequential unique whole numbers that are automatically generated. Value Data identifier created by spatial software ESRI 1 2 Count Number of pixels in that class Producer defined 165281.0 834719.0 Number of pixels in that class Red Red value in the RGB set of the color which represents this class Producer defined 0 210 Green Green value in the RGB set of the color which represents this class Producer defined 0 180 Blue Blue value in the RGB set of the color which represents this class Producer defined 140 255 Opacity Transparency Producer defined 255 255 Transparency All areas characterized as emergent vegetation, wetland forest, scrub-shrub, or uplands are classified as land, while open water aquatics, and mud flats are classified as water. Items within the attribute table in addition to ArcInfo items (e.g. area. perimeter) include: 1) Class- classified as either land or water, 2) Acres- acreage figures Producer defined U.S. Geological Survey - ScienceBase mailing and physical

Denver Federal Center, Building 810, Mail Stop 302

Denver CO 80225 USA 1-888-275-8747 sciencebase@usgs.gov Downloadable data Although these data have been processed successfully on a computer system at the U.S. Geological Survey, Wetland and Aquatic Research Center, no warranty expressed or implied is made regarding the accuracy or utility of the data on any other system or for general or scientific purposes, nor shall the act of distribution constitute any such warranty. This disclaimer applies both to individual use of the data and aggregate use with other data. It is strongly recommended that these data are directly acquired from a U.S. Geological Survey server, and not indirectly through other sources which may have changed the data in some way. It is also strongly recommended that careful attention be paid to the contents of the metadata file associated with these data. The U.S. Geological Survey shall not be held liable for improper or incorrect use of the data described and/or contained herein. Digital Data https://doi.org/10.5066/P90RE64M None None 20200830 U.S. Geological Survey, Wetland and Aquatic Research Center Brady Couvillion Research Geographer mailing and physical address

Parker Coliseum, LSU

Baton Rouge LA 70803 USA 225-578-7484 couvillionb@usgs.gov FGDC Content Standard for Digital Geospatial Metadata FGDC-STD-001-1998 local time None Acknowledgement of the U.S. Geological Survey, Wetland and Aquatic Research Center as the metadata source would be appreciated. Please cite the original metadata when using portions of the record when creating a similar record for slightly altered data for reprojection or subsetting. Any use of trade, product, or firm names is for descriptive purposes only and does not imply endorsement by the U.S. Government. http://www.esri.com/metadata/esriprof80.html ESRI Metadata Profile